Optimization of Yield and Conversion Rates in Methane Dry Reforming Using Artificial Neural Networks and the Multiobjective Genetic Algorithm

Abstract

This research aimed to enhance the dry reforming of methane by integrating computational fluid dynamics (CFD), artificial neural network (ANN), and multiobjective genetic algorithm (MOGA) techniques. Through a comprehensive analysis of reactor setups using computational fluid dynamics, reliable data were generated. Machine learning models based on the ANN were then trained with this data to establish connections between the yield, the conversion rates, the flow rates, the carbon content, and the input parameters. The trained ANN models were subsequently employed as an objective function for the MOGA, enabling the identification of optimized input parameter values that maximize the conversion rates, yields, and flow rates and minimize the carbon content. The optimization results from the MOGA revealed that in the low flow rate regime, the optimal input parameters for the reactor performance fell within a U g range of 0.08−1.0 m/ s with corresponding h/H values of 0.37 and 0.48. In the high flow rate regime, the Pareto front analysis unveiled that the optimal U g values ranged from 1.5 to 1.6 m/s accompanied by corresponding h/H values ranging from 0.5 to 0.8. Furthermore, the study identified temperature values between 814 and 817 m/s as the optimal range for achieving a balanced compromise between the coke content and the conversion rates/yields. Overall, this research provides valuable insights into optimizing the dry reforming process, highlighting the effectiveness of the ANN models, the significance of specialized methods for optimization, and the relationship between the input parameters and the performance indicators. These findings contribute to the development of more efficient and productive reactors for the dry reforming of methane.

Recent Advances in Coke Management for Dry Reforming of Methane over Ni-Based Catalysts

“…Therefore, an advanced catalyst design should consider various limiting factors, be suitable for more severe reaction conditions, and reduce reactant requirements. Using intelligent technologies (e.g., machine learning [232], artificial intelligence [233]), and density functional theory (DFT) [234] to assist catalyst design is a promising approach. For instance, descriptors obtained through DFT or experimental data play a crucial role in efficiently screening massive catalytic materials for the DRM, and machine learning and artificial neural networks can predict key catalytic performance indicators, such as CH 4 and CO 2 conversion, the synthesis gas ratio, and carbon deposition, thus contributing to the reduction of manual costs, acceleration of the catalyst development processes, and promotion of the widespread application of catalyst technology under more severe reaction conditions.…”

Review and Outlook of Confined Ni Catalysts for the Dry Reforming of Methane Reaction

“…Considering the stricter physical and chemical confinement effects in the confined catalysts, Ni sintering and carbon deposition may be more alleviated under fields of light, microwave, and electricity for DRM. (3) Besides the above experimental studies, the theoretical studies, such as functional theory (DFT) calculation, artificial neural network, and reactor modeling, could be considered for DRM over the confined catalysts. − The DFT calculation can reveal fundamental knowledge on DRM and can guide the design of core–shell catalysts, which help pioneers greatly understand the reaction pathways and shorten the exploring period of confined catalysts . The recently rising technology of artificial intelligence and machine leaning can promote the design and choose of catalysts for DRM, which can predict performance data and optimize reaction condition parameters through the computational calculation without real catalyst preparation and DRM performance testing .…”

Multiphase Transportation, Conversion, & Utilization of Energy in Chemical Engineering: A Special Issue for MTCUE-2022

“…The results demonstrated that gas–liquid contact time of the optimized unit increased by more than three times compared to the base unit, while energy consumption was reduced by 85%. Based on the CFD research, Berrouk et al combined deep neural network and multiobjective genetic algorithm techniques to perform a comprehensive analysis of the reactor setups, which enabled determination of optimized input parameter values to maximize conversion rates, yields, flow rates and minimize carbon content. The optimum U_g range of 0.08 to 1.0 m/s, with corresponding h/H values of 0.37 and 0.48, was found to be the best input parameter for reactor performance at low flow conditions.…”